Re: Bottom up approach to web services

On Tuesday, September 9, 2003, at 01:12 AM, charlie@semantech.org wrote:

>
> I have checked that out, thanks. To what extent can one automate such
> composition process?

To varying extents, I'd guess. In part, our composer provides some 
(assisted) automation. Of course, it's only composing sequences (one 
with more complex control constructs on the way, and see sheila's) 
We're investigating various ways to extend the *kind* of automation as 
well.

>  Also it seems to me that not being able to make use of the
> control constructs in the final composition is restrictive and does 
> not take
> advantage of the expressive potential of the process model.

Well, it is restrictive, no doubt. But it's also surprisingly useful :)

Our work with the SHOP planner also generates only sequences, but 
that's something of an artifact. It's perfectly possible to generate 
conditional plans (for example). In shop, you could record the 
decisions (or some decisions) made in the task network. That's easy 
enough. So, one can have fully automated generatiion of DAML-S 
processes with interesting control constructs (I'm sure other planning 
methodologies or systems could do some of this, but I'm not up on them 
at the moment).

However, it might be less exciting that you might hope for. It's not 
magic. It's not like the planner can magically write interesting 
programs to do interesting things that you, the user, didn't even know 
you wanted to see done. For an HTN planner, you need a considerable 
body of methods for the planner to work against, and some substantial 
set of them have to have been hand crafted (feeding back the results of 
the planner, even with conditional plans, can only increase the 
efficiently of finding a solution, not increase the solution space).

Hope this helps.

Cheers,
Bijan Parsia.

Received on Tuesday, 9 September 2003 09:02:05 UTC